RuntimeError: length >= offset INTERNAL ASSERT FAILED at "/Users/runner/work/_temp/anaconda/conda-bld/pytorch_1675757334591/work/aten/src/ATen/mps/MPSStream.mm":122, please report a bug to PyTorch. #94390
Description
🐛 Describe the bug
The following code deterministically produces the following output and error when running on MPS, but not on CPU:
0
1
2
Traceback (most recent call last):
File "/Users/ivan/Documents/Code/Year1/IndependentProject/test_bug.py", line 9, in <module>
interp_z[:, i] = interpolation[i]
RuntimeError: length >= offset INTERNAL ASSERT FAILED at "/Users/runner/work/_temp/anaconda/conda-bld/pytorch_1675757334591/work/aten/src/ATen/mps/MPSStream.mm":122, please report a bug to PyTorch.
Code:
import torch
z = torch.rand((1, 10))
interpolation = torch.zeros(10)
for i in range(z.size(1)):
print(i)
interp_z = z.clone().to("mps") # does not error when .to("cpu")
interp_z[:, i] = interpolation[i]Versions
Collecting environment information...
PyTorch version: 2.0.0.dev20230207
Is debug build: False
CUDA used to build PyTorch: None
ROCM used to build PyTorch: N/A
OS: macOS 12.6 (arm64)
GCC version: Could not collect
Clang version: 14.0.0 (clang-1400.0.29.102)
CMake version: Could not collect
Libc version: N/A
Python version: 3.9.16 | packaged by conda-forge | (main, Feb 1 2023, 21:38:11) [Clang 14.0.6 ] (64-bit runtime)
Python platform: macOS-12.6-arm64-arm-64bit
Is CUDA available: False
CUDA runtime version: No CUDA
CUDA_MODULE_LOADING set to: N/A
GPU models and configuration: No CUDA
Nvidia driver version: No CUDA
cuDNN version: No CUDA
HIP runtime version: N/A
MIOpen runtime version: N/A
Is XNNPACK available: True
CPU:
Apple M2
Versions of relevant libraries:
[pip3] mypy-extensions==1.0.0
[pip3] numpy==1.24.2
[pip3] pytorch-lightning==1.9.0
[pip3] torch==2.0.0.dev20230207
[pip3] torchmetrics==0.11.1
[pip3] torchvision==0.15.0.dev20230207
[conda] numpy 1.24.2 py39hff61c6a_0 conda-forge
[conda] pytorch 2.0.0.dev20230207 py3.9_0 pytorch-nightly
[conda] pytorch-lightning 1.9.0 pyhd8ed1ab_1 conda-forge
[conda] torchmetrics 0.11.1 pyhd8ed1ab_0 conda-forge
[conda] torchvision 0.15.0.dev20230207 py39_cpu pytorch-nightly
cc @kulinseth @albanD @malfet @DenisVieriu97 @razarmehr @abhudev